Home | History | Annotate | Download | only in cpu_ref
      1 /*
      2  * Copyright (C) 2011-2012 The Android Open Source Project
      3  *
      4  * Licensed under the Apache License, Version 2.0 (the "License");
      5  * you may not use this file except in compliance with the License.
      6  * You may obtain a copy of the License at
      7  *
      8  *      http://www.apache.org/licenses/LICENSE-2.0
      9  *
     10  * Unless required by applicable law or agreed to in writing, software
     11  * distributed under the License is distributed on an "AS IS" BASIS,
     12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
     13  * See the License for the specific language governing permissions and
     14  * limitations under the License.
     15  */
     16 
     17 #include "rsContext.h"
     18 #include "rsScriptC.h"
     19 #include "rsMatrix4x4.h"
     20 #include "rsMatrix3x3.h"
     21 #include "rsMatrix2x2.h"
     22 #include "rsRuntime.h"
     23 
     24 #include "rsCpuCore.h"
     25 #include "rsCpuScript.h"
     26 
     27 #include <time.h>
     28 
     29 using namespace android;
     30 using namespace android::renderscript;
     31 
     32 typedef float float2 __attribute__((ext_vector_type(2)));
     33 typedef float float3 __attribute__((ext_vector_type(3)));
     34 typedef float float4 __attribute__((ext_vector_type(4)));
     35 typedef char char2 __attribute__((ext_vector_type(2)));
     36 typedef char char3 __attribute__((ext_vector_type(3)));
     37 typedef char char4 __attribute__((ext_vector_type(4)));
     38 typedef unsigned char uchar2 __attribute__((ext_vector_type(2)));
     39 typedef unsigned char uchar3 __attribute__((ext_vector_type(3)));
     40 typedef unsigned char uchar4 __attribute__((ext_vector_type(4)));
     41 typedef short short2 __attribute__((ext_vector_type(2)));
     42 typedef short short3 __attribute__((ext_vector_type(3)));
     43 typedef short short4 __attribute__((ext_vector_type(4)));
     44 typedef unsigned short ushort2 __attribute__((ext_vector_type(2)));
     45 typedef unsigned short ushort3 __attribute__((ext_vector_type(3)));
     46 typedef unsigned short ushort4 __attribute__((ext_vector_type(4)));
     47 typedef int32_t int2 __attribute__((ext_vector_type(2)));
     48 typedef int32_t int3 __attribute__((ext_vector_type(3)));
     49 typedef int32_t int4 __attribute__((ext_vector_type(4)));
     50 typedef uint32_t uint2 __attribute__((ext_vector_type(2)));
     51 typedef uint32_t uint3 __attribute__((ext_vector_type(3)));
     52 typedef uint32_t uint4 __attribute__((ext_vector_type(4)));
     53 typedef long long long2 __attribute__((ext_vector_type(2)));
     54 typedef long long long3 __attribute__((ext_vector_type(3)));
     55 typedef long long long4 __attribute__((ext_vector_type(4)));
     56 typedef unsigned long long ulong2 __attribute__((ext_vector_type(2)));
     57 typedef unsigned long long ulong3 __attribute__((ext_vector_type(3)));
     58 typedef unsigned long long ulong4 __attribute__((ext_vector_type(4)));
     59 
     60 
     61 //////////////////////////////////////////////////////////////////////////////
     62 // Message routines
     63 //////////////////////////////////////////////////////////////////////////////
     64 
     65 
     66 static void SC_debugF(const char *s, float f) {
     67     ALOGD("float %s %f, 0x%08x", s, f, *((int *) (&f)));
     68 }
     69 static void SC_debugFv2(const char *s, float f1, float f2) {
     70     ALOGD("float x2 %s {%f, %f}", s, f1, f2);
     71 }
     72 static void SC_debugFv3(const char *s, float f1, float f2, float f3) {
     73     ALOGD("float x3 %s {%f, %f, %f}", s, f1, f2, f3);
     74 }
     75 static void SC_debugFv4(const char *s, float f1, float f2, float f3, float f4) {
     76     ALOGD("float x4 %s {%f, %f, %f, %f}", s, f1, f2, f3, f4);
     77 }
     78 static void SC_debugF2(const char *s, const float2 *f) {
     79     ALOGD("float2 %s {%f, %f}", s, f->x, f->y);
     80 }
     81 static void SC_debugF3(const char *s, const float3 *f) {
     82     ALOGD("float3 %s {%f, %f, %f}", s, f->x, f->y, f->z);
     83 }
     84 static void SC_debugF4(const char *s, const float4 *f) {
     85     ALOGD("float4 %s {%f, %f, %f, %f}", s, f->x, f->y, f->z, f->w);
     86 }
     87 static void SC_debugD(const char *s, double d) {
     88     ALOGD("double %s %f, 0x%08llx", s, d, *((long long *) (&d)));
     89 }
     90 static void SC_debugFM4v4(const char *s, const float *f) {
     91     ALOGD("matrix4x4 %s {%f, %f, %f, %f", s, f[0], f[4], f[8], f[12]);
     92     ALOGD("          %s  %f, %f, %f, %f", s, f[1], f[5], f[9], f[13]);
     93     ALOGD("          %s  %f, %f, %f, %f", s, f[2], f[6], f[10], f[14]);
     94     ALOGD("          %s  %f, %f, %f, %f}", s, f[3], f[7], f[11], f[15]);
     95 }
     96 static void SC_debugFM3v3(const char *s, const float *f) {
     97     ALOGD("matrix3x3 %s {%f, %f, %f", s, f[0], f[3], f[6]);
     98     ALOGD("          %s  %f, %f, %f", s, f[1], f[4], f[7]);
     99     ALOGD("          %s  %f, %f, %f}",s, f[2], f[5], f[8]);
    100 }
    101 static void SC_debugFM2v2(const char *s, const float *f) {
    102     ALOGD("matrix2x2 %s {%f, %f", s, f[0], f[2]);
    103     ALOGD("          %s  %f, %f}",s, f[1], f[3]);
    104 }
    105 static void SC_debugI8(const char *s, char c) {
    106     ALOGD("char %s %hhd  0x%hhx", s, c, (unsigned char)c);
    107 }
    108 static void SC_debugC2(const char *s, const char2 *c) {
    109     ALOGD("char2 %s {%hhd, %hhd}  0x%hhx 0x%hhx", s, c->x, c->y, (unsigned char)c->x, (unsigned char)c->y);
    110 }
    111 static void SC_debugC3(const char *s, const char3 *c) {
    112     ALOGD("char3 %s {%hhd, %hhd, %hhd}  0x%hhx 0x%hhx 0x%hhx", s, c->x, c->y, c->z, (unsigned char)c->x, (unsigned char)c->y, (unsigned char)c->z);
    113 }
    114 static void SC_debugC4(const char *s, const char4 *c) {
    115     ALOGD("char4 %s {%hhd, %hhd, %hhd, %hhd}  0x%hhx 0x%hhx 0x%hhx 0x%hhx", s, c->x, c->y, c->z, c->w, (unsigned char)c->x, (unsigned char)c->y, (unsigned char)c->z, (unsigned char)c->w);
    116 }
    117 static void SC_debugU8(const char *s, unsigned char c) {
    118     ALOGD("uchar %s %hhu  0x%hhx", s, c, c);
    119 }
    120 static void SC_debugUC2(const char *s, const uchar2 *c) {
    121     ALOGD("uchar2 %s {%hhu, %hhu}  0x%hhx 0x%hhx", s, c->x, c->y, c->x, c->y);
    122 }
    123 static void SC_debugUC3(const char *s, const uchar3 *c) {
    124     ALOGD("uchar3 %s {%hhu, %hhu, %hhu}  0x%hhx 0x%hhx 0x%hhx", s, c->x, c->y, c->z, c->x, c->y, c->z);
    125 }
    126 static void SC_debugUC4(const char *s, const uchar4 *c) {
    127     ALOGD("uchar4 %s {%hhu, %hhu, %hhu, %hhu}  0x%hhx 0x%hhx 0x%hhx 0x%hhx", s, c->x, c->y, c->z, c->w, c->x, c->y, c->z, c->w);
    128 }
    129 static void SC_debugI16(const char *s, short c) {
    130     ALOGD("short %s %hd  0x%hx", s, c, c);
    131 }
    132 static void SC_debugS2(const char *s, const short2 *c) {
    133     ALOGD("short2 %s {%hd, %hd}  0x%hx 0x%hx", s, c->x, c->y, c->x, c->y);
    134 }
    135 static void SC_debugS3(const char *s, const short3 *c) {
    136     ALOGD("short3 %s {%hd, %hd, %hd}  0x%hx 0x%hx 0x%hx", s, c->x, c->y, c->z, c->x, c->y, c->z);
    137 }
    138 static void SC_debugS4(const char *s, const short4 *c) {
    139     ALOGD("short4 %s {%hd, %hd, %hd, %hd}  0x%hx 0x%hx 0x%hx 0x%hx", s, c->x, c->y, c->z, c->w, c->x, c->y, c->z, c->w);
    140 }
    141 static void SC_debugU16(const char *s, unsigned short c) {
    142     ALOGD("ushort %s %hu  0x%hx", s, c, c);
    143 }
    144 static void SC_debugUS2(const char *s, const ushort2 *c) {
    145     ALOGD("ushort2 %s {%hu, %hu}  0x%hx 0x%hx", s, c->x, c->y, c->x, c->y);
    146 }
    147 static void SC_debugUS3(const char *s, const ushort3 *c) {
    148     ALOGD("ushort3 %s {%hu, %hu, %hu}  0x%hx 0x%hx 0x%hx", s, c->x, c->y, c->z, c->x, c->y, c->z);
    149 }
    150 static void SC_debugUS4(const char *s, const ushort4 *c) {
    151     ALOGD("ushort4 %s {%hu, %hu, %hu, %hu}  0x%hx 0x%hx 0x%hx 0x%hx", s, c->x, c->y, c->z, c->w, c->x, c->y, c->z, c->w);
    152 }
    153 static void SC_debugI32(const char *s, int32_t i) {
    154     ALOGD("int %s %d  0x%x", s, i, i);
    155 }
    156 static void SC_debugI2(const char *s, const int2 *i) {
    157     ALOGD("int2 %s {%d, %d}  0x%x 0x%x", s, i->x, i->y, i->x, i->y);
    158 }
    159 static void SC_debugI3(const char *s, const int3 *i) {
    160     ALOGD("int3 %s {%d, %d, %d}  0x%x 0x%x 0x%x", s, i->x, i->y, i->z, i->x, i->y, i->z);
    161 }
    162 static void SC_debugI4(const char *s, const int4 *i) {
    163     ALOGD("int4 %s {%d, %d, %d, %d}  0x%x 0x%x 0x%x 0x%x", s, i->x, i->y, i->z, i->w, i->x, i->y, i->z, i->w);
    164 }
    165 static void SC_debugU32(const char *s, uint32_t i) {
    166     ALOGD("uint %s %u  0x%x", s, i, i);
    167 }
    168 static void SC_debugUI2(const char *s, const uint2 *i) {
    169     ALOGD("uint2 %s {%u, %u}  0x%x 0x%x", s, i->x, i->y, i->x, i->y);
    170 }
    171 static void SC_debugUI3(const char *s, const uint3 *i) {
    172     ALOGD("uint3 %s {%u, %u, %u}  0x%x 0x%x 0x%x", s, i->x, i->y, i->z, i->x, i->y, i->z);
    173 }
    174 static void SC_debugUI4(const char *s, const uint4 *i) {
    175     ALOGD("uint4 %s {%u, %u, %u, %u}  0x%x 0x%x 0x%x 0x%x", s, i->x, i->y, i->z, i->w, i->x, i->y, i->z, i->w);
    176 }
    177 static void SC_debugLL64(const char *s, long long ll) {
    178     ALOGD("long %s %lld  0x%llx", s, ll, ll);
    179 }
    180 static void SC_debugL2(const char *s, const long2 *ll) {
    181     ALOGD("long2 %s {%lld, %lld}  0x%llx 0x%llx", s, ll->x, ll->y, ll->x, ll->y);
    182 }
    183 static void SC_debugL3(const char *s, const long3 *ll) {
    184     ALOGD("long3 %s {%lld, %lld, %lld}  0x%llx 0x%llx 0x%llx", s, ll->x, ll->y, ll->z, ll->x, ll->y, ll->z);
    185 }
    186 static void SC_debugL4(const char *s, const long4 *ll) {
    187     ALOGD("long4 %s {%lld, %lld, %lld, %lld}  0x%llx 0x%llx 0x%llx 0x%llx", s, ll->x, ll->y, ll->z, ll->w, ll->x, ll->y, ll->z, ll->w);
    188 }
    189 static void SC_debugULL64(const char *s, unsigned long long ll) {
    190     ALOGD("ulong %s %llu  0x%llx", s, ll, ll);
    191 }
    192 static void SC_debugUL2(const char *s, const ulong2 *ll) {
    193     ALOGD("ulong2 %s {%llu, %llu}  0x%llx 0x%llx", s, ll->x, ll->y, ll->x, ll->y);
    194 }
    195 static void SC_debugUL3(const char *s, const ulong3 *ll) {
    196     ALOGD("ulong3 %s {%llu, %llu, %llu}  0x%llx 0x%llx 0x%llx", s, ll->x, ll->y, ll->z, ll->x, ll->y, ll->z);
    197 }
    198 static void SC_debugUL4(const char *s, const ulong4 *ll) {
    199     ALOGD("ulong4 %s {%llu, %llu, %llu, %llu}  0x%llx 0x%llx 0x%llx 0x%llx", s, ll->x, ll->y, ll->z, ll->w, ll->x, ll->y, ll->z, ll->w);
    200 }
    201 static void SC_debugP(const char *s, const void *p) {
    202     ALOGD("void * %s %p", s, p);
    203 }
    204 
    205 
    206 //////////////////////////////////////////////////////////////////////////////
    207 // Stub implementation
    208 //////////////////////////////////////////////////////////////////////////////
    209 
    210 // llvm name mangling ref
    211 //  <builtin-type> ::= v  # void
    212 //                 ::= b  # bool
    213 //                 ::= c  # char
    214 //                 ::= a  # signed char
    215 //                 ::= h  # unsigned char
    216 //                 ::= s  # short
    217 //                 ::= t  # unsigned short
    218 //                 ::= i  # int
    219 //                 ::= j  # unsigned int
    220 //                 ::= l  # long
    221 //                 ::= m  # unsigned long
    222 //                 ::= x  # long long, __int64
    223 //                 ::= y  # unsigned long long, __int64
    224 //                 ::= f  # float
    225 //                 ::= d  # double
    226 
    227 static RsdCpuReference::CpuSymbol gSyms[] = {
    228     { "memset", (void *)&memset, true },
    229     { "memcpy", (void *)&memcpy, true },
    230 
    231     // Debug
    232     { "_Z7rsDebugPKcf", (void *)&SC_debugF, true },
    233     { "_Z7rsDebugPKcff", (void *)&SC_debugFv2, true },
    234     { "_Z7rsDebugPKcfff", (void *)&SC_debugFv3, true },
    235     { "_Z7rsDebugPKcffff", (void *)&SC_debugFv4, true },
    236     { "_Z7rsDebugPKcPKDv2_f", (void *)&SC_debugF2, true },
    237     { "_Z7rsDebugPKcPKDv3_f", (void *)&SC_debugF3, true },
    238     { "_Z7rsDebugPKcPKDv4_f", (void *)&SC_debugF4, true },
    239     { "_Z7rsDebugPKcd", (void *)&SC_debugD, true },
    240     { "_Z7rsDebugPKcPK12rs_matrix4x4", (void *)&SC_debugFM4v4, true },
    241     { "_Z7rsDebugPKcPK12rs_matrix3x3", (void *)&SC_debugFM3v3, true },
    242     { "_Z7rsDebugPKcPK12rs_matrix2x2", (void *)&SC_debugFM2v2, true },
    243     { "_Z7rsDebugPKcc", (void *)&SC_debugI8, true },
    244     { "_Z7rsDebugPKcPKDv2_c", (void *)&SC_debugC2, true },
    245     { "_Z7rsDebugPKcPKDv3_c", (void *)&SC_debugC3, true },
    246     { "_Z7rsDebugPKcPKDv4_c", (void *)&SC_debugC4, true },
    247     { "_Z7rsDebugPKch", (void *)&SC_debugU8, true },
    248     { "_Z7rsDebugPKcPKDv2_h", (void *)&SC_debugUC2, true },
    249     { "_Z7rsDebugPKcPKDv3_h", (void *)&SC_debugUC3, true },
    250     { "_Z7rsDebugPKcPKDv4_h", (void *)&SC_debugUC4, true },
    251     { "_Z7rsDebugPKcs", (void *)&SC_debugI16, true },
    252     { "_Z7rsDebugPKcPKDv2_s", (void *)&SC_debugS2, true },
    253     { "_Z7rsDebugPKcPKDv3_s", (void *)&SC_debugS3, true },
    254     { "_Z7rsDebugPKcPKDv4_s", (void *)&SC_debugS4, true },
    255     { "_Z7rsDebugPKct", (void *)&SC_debugU16, true },
    256     { "_Z7rsDebugPKcPKDv2_t", (void *)&SC_debugUS2, true },
    257     { "_Z7rsDebugPKcPKDv3_t", (void *)&SC_debugUS3, true },
    258     { "_Z7rsDebugPKcPKDv4_t", (void *)&SC_debugUS4, true },
    259     { "_Z7rsDebugPKci", (void *)&SC_debugI32, true },
    260     { "_Z7rsDebugPKcPKDv2_i", (void *)&SC_debugI2, true },
    261     { "_Z7rsDebugPKcPKDv3_i", (void *)&SC_debugI3, true },
    262     { "_Z7rsDebugPKcPKDv4_i", (void *)&SC_debugI4, true },
    263     { "_Z7rsDebugPKcj", (void *)&SC_debugU32, true },
    264     { "_Z7rsDebugPKcPKDv2_j", (void *)&SC_debugUI2, true },
    265     { "_Z7rsDebugPKcPKDv3_j", (void *)&SC_debugUI3, true },
    266     { "_Z7rsDebugPKcPKDv4_j", (void *)&SC_debugUI4, true },
    267     // Both "long" and "unsigned long" need to be redirected to their
    268     // 64-bit counterparts, since we have hacked Slang to use 64-bit
    269     // for "long" on Arm (to be similar to Java).
    270     { "_Z7rsDebugPKcl", (void *)&SC_debugLL64, true },
    271     { "_Z7rsDebugPKcPKDv2_l", (void *)&SC_debugL2, true },
    272     { "_Z7rsDebugPKcPKDv3_l", (void *)&SC_debugL3, true },
    273     { "_Z7rsDebugPKcPKDv4_l", (void *)&SC_debugL4, true },
    274     { "_Z7rsDebugPKcm", (void *)&SC_debugULL64, true },
    275     { "_Z7rsDebugPKcPKDv2_m", (void *)&SC_debugUL2, true },
    276     { "_Z7rsDebugPKcPKDv3_m", (void *)&SC_debugUL3, true },
    277     { "_Z7rsDebugPKcPKDv4_m", (void *)&SC_debugUL4, true },
    278     { "_Z7rsDebugPKcx", (void *)&SC_debugLL64, true },
    279     { "_Z7rsDebugPKcPKDv2_x", (void *)&SC_debugL2, true },
    280     { "_Z7rsDebugPKcPKDv3_x", (void *)&SC_debugL3, true },
    281     { "_Z7rsDebugPKcPKDv4_x", (void *)&SC_debugL4, true },
    282     { "_Z7rsDebugPKcy", (void *)&SC_debugULL64, true },
    283     { "_Z7rsDebugPKcPKDv2_y", (void *)&SC_debugUL2, true },
    284     { "_Z7rsDebugPKcPKDv3_y", (void *)&SC_debugUL3, true },
    285     { "_Z7rsDebugPKcPKDv4_y", (void *)&SC_debugUL4, true },
    286     { "_Z7rsDebugPKcPKv", (void *)&SC_debugP, true },
    287 
    288     { NULL, NULL, false }
    289 };
    290 
    291 
    292 void * RsdCpuScriptImpl::lookupRuntimeStub(void* pContext, char const* name) {
    293     RsdCpuScriptImpl *s = (RsdCpuScriptImpl *)pContext;
    294     const RsdCpuReference::CpuSymbol *syms = gSyms;
    295     const RsdCpuReference::CpuSymbol *sym = NULL;
    296 
    297     sym = s->mCtx->symLookup(name);
    298     if (!sym) {
    299         sym = s->lookupSymbolMath(name);
    300     }
    301     if (!sym) {
    302         while (syms->fnPtr) {
    303             if (!strcmp(syms->name, name)) {
    304                 sym = syms;
    305             }
    306             syms++;
    307         }
    308     }
    309 
    310     if (sym) {
    311         s->mIsThreadable &= sym->threadable;
    312         return sym->fnPtr;
    313     }
    314     ALOGE("ScriptC sym lookup failed for %s", name);
    315     return NULL;
    316 }
    317 
    318 
    319